A method for identifying loudspeaker defects based on voice broadcast sound quality

Abstract

The present invention relates to the technical field of audio signal processing, and in particular to a method for identifying speaker defects based on voice broadcast sound quality, comprising: using a broadcast sample to broadcast in a speaker to be tested, wherein the broadcast sample is a language audio sequence; collecting a voice signal played by the speaker to be tested and preprocessing it, and obtaining a language broadcast signal after removing a noise segment; dividing the language broadcast signal and the broadcast sample into a plurality of sub-signals to obtain a broadcast sub-signal set and a sample sub-signal set; performing spectrum feature extraction, signal matching, abnormal sound detection and harmonic wave analysis on the sub-signals in each set to extract psychoacoustic features; comparing the extracted psychoacoustic features with preset standard values ​​to evaluate the sound quality performance of the speaker, so that the quality of the voice signal played by the speaker can be evaluated from a psychoacoustic perspective by analyzing the voice signal, thereby realizing automatic detection of psychoacoustic defects of the speaker.

Description

Technical Field

[0001] The present invention relates to the technical field of audio signal processing, and in particular to a method for identifying loudspeaker defects based on voice broadcast sound quality. Background Art

[0002] Speakers are the core components of audio equipment, and their quality directly affects the sound quality of audio equipment. However, in addition to physical acoustic parameters, people's perception of audio equipment is also affected by psychoacoustic factors. Psychoacoustic factors include but are not limited to the comfort, naturalness, clarity, etc. of sound, which are often not easy to describe and quantify using traditional physical parameters. At present, the detection of psychoacoustic defects of speakers mainly relies on manual subjective evaluation, which has problems such as strong subjectivity, poor consistency, and low efficiency.

[0003] Therefore, it is particularly important to develop an automated, objective method that can accurately identify loudspeaker psychoacoustic defects.

[0004] With the widespread application and increasing demand for audio equipment, the real-time requirements for speaker defect detection algorithms are becoming increasingly higher. Especially on the production line, it is necessary to be able to detect speakers quickly and accurately to ensure production efficiency and product quality.

[0005] Developing an automated loudspeaker defect detection algorithm that can overcome the problems of existing loudspeaker detection methods and has real-time performance, strong applicability, good anti-interference ability and strong adaptability has important research significance and application value. Summary of the invention

[0006] The purpose of the present invention is to provide a method for identifying loudspeaker defects based on voice broadcast sound quality, aiming to analyze the voice signal played by the loudspeaker, evaluate its quality from a psychoacoustic perspective, and realize automatic detection of loudspeaker psychoacoustic defects.

[0007] To achieve the above object, the present invention provides a method for identifying speaker defects based on voice broadcast sound quality, comprising using a broadcast sample to broadcast in a speaker to be tested, wherein the broadcast sample is a language audio sequence;

[0008] Collect the voice signal played by the speaker to be tested and pre-process it, and obtain the language broadcast signal after removing the noise segment;

[0009] Dividing the language broadcast signal and the broadcast sample into a plurality of sub-signals to obtain a broadcast sub-signal set and a sample sub-signal set;

[0010] Perform spectrum feature extraction, signal matching, noise detection and harmonic analysis on the sub-signals in each set to extract psychoacoustic features;

[0011] The extracted psychoacoustic features are compared with preset standard values ​​to evaluate the sound quality performance of the speaker;

[0012] Output sound quality performance results.

[0013] The speaker to be tested operates at rated power and avoids overload to maintain normal working state.

[0014] The speech signal played by the speaker to be tested is collected using a high-precision microphone, and the microphone has flat frequency response and low self-noise characteristics.

[0015] The specific steps of using the broadcast sample to broadcast in the speaker to be tested, wherein the broadcast sample is a language audio sequence, include:

[0016] Prepare speech samples;

[0017] Test the speaker to be tested to ensure that it works within the normal range;

[0018] Play the speech sample through the speaker under test.

[0019] The specific steps of collecting the voice signal played by the speaker to be tested and preprocessing it to obtain the language broadcast signal after removing the noise segment include:

[0020] Use a high-precision microphone to collect the voice signal played by the speaker;

[0021] Performing filtering on the collected speech signal;

[0022] The filtered speech signal is normalized to eliminate the difference in signal strength;

[0023] The signal cross-correlation algorithm is used to calculate the starting position of the signal, the signal ending position is determined according to the source signal waveform length, and the environmental noise before and after the broadcast is intercepted to obtain the language broadcast signal.

[0024] Wherein, the frequency range of the bandpass filter used for filtering the collected voice signal is 20 Hz to 20 kHz.

[0025] The specific steps of using the signal cross-correlation algorithm to calculate the signal start position, determining the signal end position according to the source signal waveform length, intercepting the environmental noise before and after the broadcast, and obtaining the language broadcast signal include:

[0026] Select a short signal segment with obvious characteristic difference from the language broadcast as the reference signal;

[0027] The correlation between the entire audio signal and the reference signal is calculated using a cross-correlation algorithm;

[0028] By analyzing the cross-correlation function graph, find the peak point;

[0029] The first significant positive peak of the cross-correlation function indicates the starting position of the speech broadcast, and the ending position of the speech broadcast is determined by adding the speech time length to the starting position;

[0030] According to the determined start and end positions, the language broadcast part is cut out from the original audio signal.

[0031] The specific steps of dividing the voice broadcast signal and the sample signal into a plurality of sub-signals include:

[0032] Use cross-correlation to detect the speech endpoints of the speech broadcast signal, determine the start and end positions of each word based on the detection results, and split the speech signal into multiple sub-signals;

[0033] The sample signal is split in the same way to ensure that the split sub-signals correspond to the voice broadcast signal in terms of time and content, and the split sub-signals and their timestamp information are saved.

[0034] The specific steps of extracting spectrum features, signal matching, noise detection and harmonic analysis on the sub-signals in each set to extract psychoacoustic features include:

[0035] Perform feature extraction on each split sub-signal to extract spectrum characteristics;

[0036] The split recorded sub-signal and the standard sample sub-signal are matched with each other using the Euclidean distance, and the similarity between the feature vectors of each pair of sub-signals is calculated;

[0037] Analyze the spectrum of each sub-signal to obtain abnormal peaks and frequency components;

[0038] Analyze the harmonic components in each sub-signal, perform Fourier transform on each sub-signal, calculate its spectrum, identify the fundamental wave and each harmonic, calculate the harmonic distortion, and save the harmonic analysis results.

[0039] The present invention provides a method for identifying speaker defects based on voice broadcast sound quality, comprising: using a broadcast sample to broadcast in the speaker to be tested, wherein the broadcast sample is a language audio sequence; collecting the voice signal played by the speaker to be tested and preprocessing it, and obtaining a language broadcast signal after removing the noise segment; dividing the language broadcast signal and the broadcast sample into multiple sub-signals to obtain a broadcast sub-signal set and a sample sub-signal set; performing spectrum feature extraction, signal matching, abnormal sound detection and harmonic analysis on the sub-signals in each set to extract psychoacoustic features; comparing the extracted psychoacoustic features with preset standard values ​​to evaluate the sound quality performance of the speaker; and outputting the sound quality performance results. Thus, it has the following advantages:

[0040] More comprehensive evaluation: The introduction of psychoacoustic factors can more comprehensively evaluate the sound quality of the speaker, starting from the user's subjective experience, which is more in line with actual usage scenarios.

[0041] Strong objectivity: Compared with the traditional subjective evaluation method, the analysis based on psychoacoustic characteristics of the present invention is more objective and avoids the differences caused by subjective evaluations of different people.

[0042] The use of automated evaluation methods can quickly and accurately identify possible psychoacoustic defects in speakers, improving the efficiency and accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 It is a schematic diagram of the hardware circuit composition of the test system of the present invention;

[0045] Figure 2 This is the waveform diagram of the standard voice broadcast sample selected this time;

[0046] Figure 3 Play audio waveform for OK speaker;

[0047] Figure 4 Play audio waveform for NG speakers;

[0048] Figure 5 Response waveform distortion performance waveform diagram

[0049] Figure 6 To display a waveform diagram in response to abnormal waveform sound;

[0050] Figure 7 The abnormal sound volume curve audible to human ears calculated by the OK speaker using this algorithm;

[0051] Figure 8 This is the audible abnormal sound volume curve of the NG speaker calculated by this algorithm;

[0052] Fig. 9 is a flow chart of a method for identifying speaker defects based on voice broadcast sound quality according to a first embodiment of the present invention;

[0053] Fig.10This is a flowchart of the first embodiment of the present invention, in which a broadcast sample is used to broadcast in a speaker to be tested, wherein the broadcast sample is a language audio sequence;

[0054] Fig.11 This is a flow chart of the first embodiment of the present invention for collecting and preprocessing the voice signal played by the speaker to be tested, and removing the noise segment to obtain the language broadcast signal;

[0055] Fig.12 It is a flowchart of the first embodiment of the present invention, which uses a signal cross-correlation algorithm to calculate the signal start position, determines the signal end position according to the source signal waveform length, intercepts the environmental noise before and after the broadcast, and obtains the language broadcast signal;

[0056] Fig.13 It is a flowchart of the first embodiment of the present invention for dividing a language broadcast signal and a broadcast sample into a plurality of sub-signals to obtain a broadcast sub-signal set and a sample sub-signal set;

[0057] Fig.14 This is a flow chart of the first embodiment of the present invention for performing spectrum feature extraction, signal matching, heterophonic detection and harmonic analysis on sub-signals in each set to extract psychoacoustic features. DETAILED DESCRIPTION

[0058] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0059] First embodiment:

[0060] See also Figure 1 to Figure 14 The present invention provides a method for identifying speaker defects based on voice broadcast sound quality, comprising:

[0061] S101 uses a broadcast sample to broadcast in the speaker to be tested, wherein the broadcast sample is a language audio sequence;

[0062] The speaker to be tested operates at rated power and avoids overload to maintain normal working state.

[0063] The specific steps include:

[0064] S201 prepares a speech sample;

[0065] S202 detects the speaker to be tested to ensure that the speaker to be tested operates within a normal range;

[0066] S203 plays the voice sample through the speaker to be tested. According to Figure 1, the test system is built. The speaker-based voice broadcast psychoacoustic defect detection system includes the following:

[0067] Voice sample preparation: Check the sound quality of the sample to ensure it is free of distortion, background noise, and has high fidelity.

[0068] Speaker broadcasting: Connect the speaker to the audio source device, make sure the connection is stable, make sure the speaker works within the rated power range, broadcast in an echo-free and low-noise environment. It is recommended to use an anechoic room to avoid overload, warm up the speaker before playing the sample, and play white noise for a few minutes to reach a stable state.

[0069] S102 collects the voice signal played by the speaker to be tested and pre-processes it to obtain a language broadcast signal after removing the noise segment;

[0070] The specific steps include:

[0071] S301 uses a high-precision microphone to collect the voice signal played by the speaker;

[0072] Choose a high-precision microphone with a flat frequency response and low self-noise, place the microphone at a suitable distance from the speaker (usually 5cm) to ensure the sound pickup effect, calibrate the microphone to ensure it is in the best working condition, power the microphone, use a professional audio interface device for acquisition, and set the appropriate sampling rate (such as 44.1kHz or 96kHz) and bit depth (such as 16 bits).

[0073] S302 performs filtering processing on the collected voice signal;

[0074] The collected speech signal is bandpass filtered, the filter type is Chebyshev Filter with a bandwidth of 20-20KHz.

[0075] The transfer function formula of the bandpass filter is:

[0076] in, is the center frequency, is the bandwidth. Symbol Represents the imaginary unit, imaginary unit represents the phase change in the complex domain. The transfer function of a bandpass filter contains imaginary units in order to describe the complex representation of the phase change and amplitude change in the frequency response.

[0077] S303 normalizes the filtered speech signal to eliminate differences in signal strength;

[0078] The filtered speech signal Perform normalization processing, calculate the maximum amplitude of the signal, and adjust the signal range to the standard range. Usually, the maximum amplitude is adjusted to 1 to ensure that there is no signal distortion during the normalization process.

[0079] The signal normalization formula is:

[0080]

[0081] S304 uses a signal cross-correlation algorithm to calculate the signal start position, determines the signal end position according to the source signal waveform length, intercepts the environmental noise before and after the broadcast, and obtains the language broadcast signal.

[0082] The environmental noise before and after the speech broadcast is cut off, the signal starting position is calculated using the signal cross-correlation algorithm, the signal ending position is determined according to the source signal waveform length, and the middle speech broadcast part is retained for subsequent analysis.

[0083] The specific steps include:

[0084] S401 selects a short segment signal having a significant characteristic difference from the language broadcast as a reference signal;

[0085] S402 uses a cross-correlation algorithm to calculate the correlation between the entire audio signal and the reference signal;

[0086] The cross-correlation function calculation formula is:

[0087]

[0088] in, is the speech signal to be detected, is the standard template signal, For time delay

[0089] S403 searches for peak points by analyzing the cross-correlation function graph;

[0090] S404: The first significant positive peak of the cross-correlation function indicates the starting position of the speech broadcast, and the ending position of the speech broadcast is determined by adding the speech time length to the starting position;

[0091] S405 extracts the language broadcast portion from the original audio signal according to the determined start and end positions.

[0092] S103: dividing the language broadcast signal and the broadcast sample into a plurality of sub-signals to obtain a broadcast sub-signal set and a sample sub-signal set;

[0093] The specific steps include:

[0094] S501 uses cross-correlation to perform voice endpoint detection of the voice broadcast signal, determines the start and end positions of each word according to the detection result, and splits the voice signal into multiple sub-signals;

[0095] The speech broadcast part after the noise is cut off is split into multiple sub-signals word by word, and the voice endpoint detection is performed using cross-correlation. The start and end positions of each word are determined based on the detection results, the speech signal is split into multiple sub-signals, and the timestamp information of each sub-signal is saved.

[0096] S502 performs the same splitting process on the sample signal to ensure that the split sub-signals correspond to the voice broadcast signal in terms of time and content, and save the split sub-signals and their time stamp information.

[0097] S104 performs spectrum feature extraction, signal matching, noise detection and harmonic analysis on the sub-signals in each set to extract psychoacoustic features;

[0098] The specific steps include:

[0099] S601 performs feature extraction on each of the split sub-signals to extract spectrum characteristics;

[0100] For each sub-signal after splitting Perform feature extraction, extract spectral characteristics, Mel-frequency cepstral coefficients (MFCC), extract time domain characteristics, short-time energy, zero-crossing rate, tone, pitch, and save the feature vector of each sub-signal for subsequent analysis.

[0101] The Mel Frequency Cepstral Coefficient (MFCC) formula is:

[0102]

[0103] Feature extraction is achieved by analyzing each sub-signal after the splitting, extracting various indicators that can reflect the psychoacoustic characteristics, including spectral characteristics, time domain characteristics, short-time energy, zero-crossing rate, tone, timbre, and pitch. The signal is converted from the time domain to the frequency domain using Fourier transform, and the spectral density and frequency components are analyzed. The signal is divided into frames and the MFCC of each frame is calculated to capture the timbre characteristics of the speech signal, including short-time energy, zero-crossing rate, tone, pitch, etc. The instantaneous characteristics of the signal are extracted through time domain analysis methods.

[0104] S602 performs feature matching between the split recorded sub-signals and the standard sample sub-signals using the Euclidean distance, and calculates the similarity between the feature vectors of each pair of sub-signals;

[0105] Match the split recorded sub-signal with the standard sample sub-signal, use Euclidean distance or dynamic time warping (DTW) method for feature matching, calculate the similarity between each pair of sub-signal feature vectors, and save the matching results.

[0106] In feature matching, the recorded sub-signal features are matched with the features of the standard sample and compared using the Euclidean distance or dynamic time warping (DTW) method.

[0107] Formula Euclidean distance:

[0108] ;

[0109] Dynamic Time Warping (DTW):

[0110] ;

[0111] The feature normalization error formula is:

[0112] .

[0113] The extracted feature vectors are normalized to eliminate the dimension difference. The similarity between the feature vectors of the recorded sub-signal and the standard sample sub-signal is calculated using the Euclidean distance or dynamic time warping (DTW) method. The calculated similarity is compared with the preset threshold to quantify the difference between the signals. The greater the difference, the more obvious the psychoacoustic defect.

[0114] S603 analyzes the frequency spectrum of each sub-signal to obtain abnormal peaks and frequency components;

[0115] Detects the presence of abnormal sounds in each sub-signal, analyzes the spectrogram of each sub-signal, looks for abnormal peaks and frequency components, uses psychoacoustic models to identify abnormal sounds such as distortion, crackling, and popping, and marks and records the detected abnormal sounds, including their location and type

[0116] S604 analyzes the harmonic components in each sub-signal, performs Fourier transform on each sub-signal, calculates its spectrum, identifies the fundamental wave and each harmonic, calculates the harmonic distortion, and saves the harmonic analysis result.

[0117] Analyze the harmonic components in each sub-signal, perform Fourier transform on each sub-signal, calculate its spectrum, identify the fundamental wave and each harmonic, calculate the harmonic distortion, and save the harmonic analysis results.

[0118] The formula for calculating total harmonic distortion (THD) is:

[0119]

[0120] in, is the fundamental voltage, , ,..., is the harmonic voltage.

[0121] The Fourier transform formula is: .

[0122] S105 compares the extracted psychoacoustic features with preset standard values ​​to evaluate the sound quality performance of the speaker;

[0123] The extracted psychoacoustic features are compared with the preset standard values, the differences between the signals are quantified, the sound quality of the speakers is evaluated, and for signals with qualified feature matching, the saved abnormal sounds and harmonics are subjected to psychoacoustic analysis. The analysis results are compared with the audible threshold of the human ear, and those above the hearing threshold of the human ear are saved.

[0124] By analyzing the shape of the spectrum, such as flatness and tilt, the clarity and naturalness of the sound can be reflected. The energy of the signal in each time period can be used to reflect the intensity and loudness of the sound. The number of times the signal crosses the zero point can be used to reflect the clarity and naturalness of the sound. The signal is pre-emphasized and framed using a Mel filter bank. The MFCC coefficients of each frame are calculated to capture the tone, timbre, and pitch characteristics of the speech signal. The nonlinear distortion components in the spectrum are analyzed to identify and locate the distortion source. The psychoacoustic model and masking effect model are used to detect the crackling and popping parts that are sensitive to the human ear. Fourier transform is performed to identify the fundamental wave and each harmonic in the signal. The harmonic distortion (THD) is calculated to determine the linear characteristics of the signal. The split recorded sub-signal is matched with the standard sample sub-signal. The differences in their spectral characteristics and time domain characteristics are compared to quantify the differences between the signals. The greater the difference, the more obvious the psychoacoustic defects.

[0125] Formulas used in psychoacoustic analysis:

[0126] Masking effect model: ;

[0127] in is the frequency The noise masking threshold at is the frequency The hearing threshold of is the reference frequency, 1kHz is used here;

[0128] Human ear audible threshold: ;

[0129] in, is the frequency The hearing threshold at and is a constant, and its specific value depends on different frequency bands;

[0130] Zwicker Loudness Model: Overall Loudness Formula

[0131] in The formula for calculating loudness for a single critical band is: , is the reference energy, The energy within the zth critical frequency band.

[0132] Because clarity is an important indicator for judging sound quality, the present invention uses clarity calculation to represent the sound quality.

[0133] Glasberg and Moore's clarity model: ; The Clarity Model proposed by Glasberg and Moore is a computational model for evaluating speech clarity in auditory scenarios. This model focuses specifically on the intelligibility of speech in noisy environments, which is very important for fields such as hearing science, hearing aid design, audio engineering, and voice communication. Glasberg and Moore's clarity model takes into account the frequency resolution of the human ear as well as the masking effect, that is, how background noise affects our perception of speech signals. The model estimates the intelligibility of speech in noise by calculating the energy contrast between speech and noise in each auditory frequency channel. Specifically, the model calculates the so-called clarity index (C), which is a value between 0 and 1, where 1 indicates perfect clarity and 0 indicates complete incomprehensibility.

[0134] in, It is The loudness of the critical bands, is a weighting function of clarity.

[0135] S106 outputs the sound quality performance result.

[0136] Use charts to intuitively display the test results.

[0137] The present invention evaluates loudspeakers based on psychoacoustic factors, which has the following advantages over the traditional evaluation method based on physical acoustic parameters:

[0138] More comprehensive evaluation: The introduction of psychoacoustic factors can more comprehensively evaluate the sound quality of the speaker, starting from the user's subjective experience, which is more in line with the actual usage scenario

[0139] Strong objectivity: Compared with the traditional subjective evaluation method, the analysis based on psychoacoustic characteristics of the present invention is more objective and avoids the differences caused by subjective evaluations of different people.

[0140] High efficiency: The automated evaluation method can quickly and accurately identify possible psychoacoustic defects of the loudspeaker, thus improving the efficiency and accuracy of detection.

[0141] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A method for identifying speaker defects based on voice broadcast sound quality, It is characterized in that include: Using a broadcast sample to broadcast in the speaker to be tested, the broadcast sample is a language audio sequence; Collect the voice signal played by the speaker to be tested and pre-process it, and obtain the language broadcast signal after removing the noise segment; The language broadcast signal and the broadcast sample are divided into a plurality of sub-signals to obtain a broadcast sub-signal set and a sample sub-signal set. The specific steps include: Use cross-correlation to detect the speech endpoints of the speech broadcast signal, determine the start and end positions of each word based on the detection results, and split the speech signal into multiple sub-signals; The sample signal is split in the same manner to ensure that the split sub-signals correspond to the voice broadcast signal in terms of time and content, and the split sub-signals and their timestamp information are saved; Perform spectrum feature extraction, signal matching, noise detection and harmonic analysis on the sub-signals in each set to extract psychoacoustic features; The extracted psychoacoustic features are compared with preset standard values ​​to evaluate the sound quality performance of the speaker; Output sound quality performance results; The specific steps of collecting the voice signal played by the speaker to be tested and preprocessing it to obtain the language broadcast signal after removing the noise segment include: Use a high-precision microphone to collect the voice signal played by the speaker; Performing filtering on the collected speech signal; The filtered speech signal is normalized to eliminate the difference in signal strength; The signal starting position is calculated using a signal cross-correlation algorithm, the signal ending position is determined according to the source signal waveform length, and the environmental noise before and after the broadcast is intercepted to obtain a language broadcast signal; the specific steps of calculating the signal starting position using a signal cross-correlation algorithm, determining the signal ending position according to the source signal waveform length, intercepting the environmental noise before and after the broadcast, and obtaining the language broadcast signal include: Select a short signal segment with obvious characteristic difference from the language broadcast as the reference signal; The correlation between the entire audio signal and the reference signal is calculated using a cross-correlation algorithm; By analyzing the cross-correlation function graph, find the peak point; The first significant positive peak of the cross-correlation function indicates the starting position of the speech broadcast, and the ending position of the speech broadcast is determined by adding the speech time length to the starting position; According to the determined start and end positions, the language broadcast part is cut out from the original audio signal.

2. A method for identifying speaker defects based on voice broadcast sound quality as claimed in claim 1, characterized in that: The speaker to be tested operates at rated power and avoids overload to maintain normal working state.

3. A method for identifying speaker defects based on voice broadcast sound quality as claimed in claim 2, characterized in that: The speech signal played by the speaker to be tested is collected using a high-precision microphone, and the microphone has flat frequency response and low noise characteristics.

4. The method for identifying speaker defects based on voice broadcast sound quality according to claim 1, characterized in that: The specific steps of using a broadcast sample to broadcast in the speaker to be tested, wherein the broadcast sample is a language audio sequence, include: Prepare speech samples; Test the speaker to be tested to ensure that it works within the normal range; Play the speech sample through the speaker under test.

5. A method for identifying speaker defects based on voice broadcast sound quality as claimed in claim 4, characterized in that: The frequency range of the bandpass filter used to filter the collected speech signal is 20 Hz to 20 kHz.

6. A method for identifying speaker defects based on voice broadcast sound quality as claimed in claim 5, characterized in that: The specific steps of extracting spectrum features, signal matching, noise detection and harmonic analysis on the sub-signals in each set to extract psychoacoustic features include: Perform feature extraction on each split sub-signal to extract spectrum characteristics; The split recorded sub-signal and the standard sample sub-signal are matched with each other using the Euclidean distance, and the similarity between the feature vectors of each pair of sub-signals is calculated; Analyze the spectrum of each sub-signal to obtain abnormal peaks and frequency components; Analyze the harmonic components in each sub-signal, perform Fourier transform on each sub-signal, calculate its spectrum, identify the fundamental wave and each harmonic, calculate the harmonic distortion, and save the harmonic analysis results.

Images